Professional translation: what separates it from AI output

Professional website translation passes four checks AI alone cannot. First, a native speaker of the target market reads every page before it ships and rewrites anything that scans as foreign. Second, the translator understands ecommerce conversion — the trust signals, CTA verbs and headline patterns that work in Germany versus France versus Italy. Third, a brand glossary is maintained so your product names, tone words and forbidden translations stay consistent across every page and every quarter. Fourth, conversion-critical copy — hero headlines, primary CTAs, marketing emails — is written human-first rather than AI-first, because that 20% of the site drives most of the revenue. AI alone passes none of these checks. Hybrid translation, where AI handles body copy at scale and native humans review every page and write the persuasive surfaces, passes all four.

Pure AI translation is not good enough for ecommerce. Published data shows AI-only translation deployed without native human review converts 25-35% lower than equivalent native-reviewed pages on conversion-critical surfaces. That is not a quality nuance — that is a third of your revenue gone before you see the analytics. The reason is structural: AI scores around 85% quality on simple body copy but collapses to roughly 40% on persuasive copy like hero headlines, CTAs and marketing emails. AI has no native intuition for what makes a buyer in Munich click versus a buyer in Lyon. It picks the safe word, and the safe word does not convert. Hybrid translation — AI plus native human review on every page, plus human-first writing on the conversion-critical 20% — recovers that gap. We use it on every site we ship.

Brand voice is locked into a maintained glossary on day one. Before any translation begins, we capture your product names, service descriptions, tone words, forbidden translations and signature phrases. Every translator and every AI pass references that glossary. When you launch a new product, the glossary updates. When a phrase stops landing in the German market and we test an alternative, the glossary updates. Quarterly we run a native review on production pages to catch drift — phrases that once worked but feel dated, or new market vocabulary worth adopting. The result is a translated site that sounds like one company speaking one consistent voice across every page, every market and every quarter, rather than a patchwork of disconnected translations done by different hands at different times.

DeepL handles European nuance — German, French, Italian, Spanish — better than any other engine and it is our default for those languages. Claude handles marketing copy where tone and persuasion matter more than literal accuracy. We use both depending on the surface. We never use Google Translate for production translation. Google Translate is fine for understanding what a foreign-language email says. It is not fine for shipping a £100k-revenue website. Its output is technically functional, conversion-poor, and immediately recognisable to native readers as machine output. The choice between DeepL and Claude is per-section, not per-site. Body copy and product descriptions go to DeepL. Hero headlines, CTAs and marketing emails go through Claude with native human refinement on top. Both AI passes are then native-reviewed before publication.

Every translation we ship comes with a customer-reported error fix SLA. On Growth and Scale tiers, errors flagged by your customers or your team are corrected and live within 48 hours. On Starter tier, the SLA is one week. Revenue-impacting errors — wrong price, broken CTA, mistranslated checkout copy, regulatory compliance issues — are same-day on every tier, including Starter. Beyond reactive fixes, we run a quarterly native review on production pages: a native speaker reads every live page, flags drift, dated phrasing or anything that has stopped landing, and we update. Translation is not a one-shot project — markets evolve, vocabulary shifts, new product lines need integration. Quarterly review keeps your translated site current rather than slowly aging into a foreign-feeling artefact six months after launch.